Prediction of mustard yield using different machine learning techniques: a case study of Rajasthan, India

Mustard is the second most important edible oilseed after groundnut for India. Adverse weather drastically reduces the mustard yield. Weather variables affect the crop differently during different stages of development. Weather influence on crop yield depends not only on the magnitude of weather variables but also on weather distribution pattern over the crop growing period. Hence, developing models using weather variables for accurate and timely crop yield prediction is foremost important for crop management and planning decisions regarding storage, import, export, etc. Machine learning plays a significant role as it has a decision support tool for crop yield prediction. The models for mustard yield prediction was developed using long-term weather data during the crop growing period along with mustard yield data. Techniques used for developing the model were variable selection using stepwise multiple linear regression (SMLR) and artificial neural network (SMLR-ANN), variable selection using SMLR and support vector machine (SMLR-SVM), variable selection using SMLR and random forest (SMLR-RF), variable extraction using principal component analysis (PCA) and ANN (PCA-ANN), variable extraction using PCA and SVM (PCA-SVM), and variable extraction using PCA and RF (PCA-RF). Optimal combinations of the developed models were done for improving the accuracy of mustard yield prediction. Results showed that, on the basis of model accuracy parameters nRMSE, RMSE, and RPD, the PCA-SVM model performed best among all the six models developed for mustard yield prediction of study areas. Performance of mustard yield prediction done by optimum combinations of the models was better than the individual model.